Re: [PATCH v2 -tip] x86/percpu: Use C for arch_raw_cpu_ptr()

[Peter Zijlstra <peterz@infradead.org>]

On Wed, Oct 18, 2023 at 03:40:05PM -0700, Linus Torvalds wrote:

> Side note: the code that caused that problem is this:
> 
>   __always_inline void __cyc2ns_read(struct cyc2ns_data *data)
>   {
>         int seq, idx;
> 
>         do {
>                 seq = this_cpu_read(cyc2ns.seq.seqcount.sequence);
>                 ...
>         } while (unlikely(seq != this_cpu_read(cyc2ns.seq.seqcount.sequence)));
>   }
> 
> where the issue is that the this_cpu_read() of that sequence number
> needs to be ordered.

I have very vague memories of other code also relying on this_cpu_read()
implying READ_ONCE().

And that code really only is buggy if you do not have that. Since it is
cpu local, the smp_rmb() would be confusing, as would smp_load_acquire()
be -- there is no cross-cpu data ordering.

The other option is of couse adding explicit barrier(), but that's
entirely superfluous when all the loads are READ_ONCE().


If you want to make this_cpu_read() not imply READ_ONCE(), then we
should go audit all users :/ Can be done ofc.